Development of Humanoid Robots in HUBO Laboratory, KAIST 369

日本ロボット学会誌 Vol. 30 No. 4, pp.367～371, 2012 367

解説 Development of Humanoid Robots in HUBO Labora- tory, KAIST

Jung-Woo Heo∗,In-HoLeeand Jun-Ho Oh ∗School of Mechanical, Aerospace & Systems Engineering, Division of Mechanical Engineering, KAIST

improvement of walking performance. We explored on- 1. Introduction line walking pattern generation [9] [10] and walking algo- The HUBO Laboratory in KAIST in the Republic rithms for uneven terrain [16]. We constantly struggled of Korea was established in 2005. Previously, this re- to upgrade the performance of the platform. From the search center was a machine control laboratory that results of this work, HUBO2 was developed in 2009. performed research on motor control. In 2000, based In this platform, running algorithm and stretched leg on motor control technologies, we began to study hu- walking is applied [17]～[22]. Recently, we developed manoid robotics in earnest. It was a very challenging HUBO2++ considering users’ convenience and com- topic at that time. Although many other research in- pleting the platform. stitutes were carrying out bipedal robot research, there This paper is organized as follows. Section 2 chrono- were few humanoid robot platforms in the world except logically describes humanoid robots developed in the ASIMO of HONDA. We were pioneers in the develop- HUBO Laboratory over 10 years. The specifications of ment of humanoid robots. each robot are shown. In section 3, the walking and Without any experience in humanoid robotics, we control algorithm of the HUBO series is reviewed. This started from a very simple and basic experiment. In review is a summary of the algorithm, not a proposal of 2000, we designed KHR-0, which was, not a humanoid specific technology. robot but just a biped walking robot without an up- 2. Humanoid Robots per body. Through this experimental platform, we de- termined the basic walking pattern and a fundamental A. KHR-1 (2003) understanding of bipedal walking. KHR-1 was the first version of a humanoid robot in Since that time, we succeeded in developing a hu- the HUBO Laboratory. It was developed in 2002 and manoid robot. In 2003, we developed KHR-1 [1]～[3], was mainly used as an experimental walking platform which was the first humanoid robot platform of the [1]. It didn’t have a head, hands or case. Although this HUBO Laboratory. In 2004, KHR-2 [4]～[7], a hu- version was made at the very beginning of humanoid manoid robot with head and hands was developed. In research in the HUBO Laboratory, it was able to walk, 2004, we developed HUBO (KHR-3) [8]～[10], a widely turn around and balance itself using a 2-axis F/T sen- known humanoid robot in the Republic of Korea. sor and a 2-axis IMU [2] [3]. These sensors were used in In 2005, we developed various kinds of bipedal hu- later versions of the humanoid robot in the HUBO Lab- manoid robots including the android-type humanoid oratory. The system configuration was almost the same robot, Albert HUBO [11] [12]. A giant human -sized as that used in later robots, but the system operated in bipedal robot, FX-1, was also developed that year [13]～ DOS. [15]. B. KHR-2 (2004) Since 2006, we extended our research scope to the KHR-2 was developed in 2003. This could be called a complete humanoid robot because it had features of 原 2012 年 3 月 26 日キーワード： Humanoid Robot, HUBO, Humanoid Robot a humanoid robot, such as a covering case, hands, a Walking, Humanoid Robot Design, Bipedal Robot head, visual cameras, etc. Further, using battery and ∗373-1 Guseong-dong, Yuseong-gu, Daejeon 305-701, Republic of Korea wireless LAN, it could walk by itself without any wires.

日本ロボット学会誌 30 巻 4 号 —33— 2012 年 5 月 368 Jung-Woo Heo In-Ho Lee Jun-Ho Oh

dently. Moreover, it can walk faster than the previous KHR series and can also climb stairs. Its self-balancing or other walking control technique was improved sig- nificantly. On the other hand, to represent the spirit of Korea, its covering case was designed to look like a Taekwondo player. Finally, after HUBO was developed, it was introduced in the worldwide media, and it has established a remarkable reputation from people Fig. 1 KHR-1 and specifications [1] the world over. Nowadays, HUBO is the most famous humanoid robot in the Republic of Korea and also rep- resents robot technology of the Republic of Korea to the world. D. Albert HUBO (2005) Albert HUBO is the first biped robot in the world that has an expressive human face. This robot was developed in 2005. The Albert HUBO adopted the techniques of the HUBO design for the body and the techniques of the Hanson Robotics for the head. It can generate a full range of facial expressions, such as a laugh, sadness, anger, surprise, etc. with ‘Fubber’ materials for smooth Fig. 2 Overall system configuration of KHR-2, HUBO, FX-1 and HUBO 2 artificial skin and 28 servo motors. It can also generate human-like motions like dynamic walking, which were introduced with HUBO (KHR-3). These robots were shown at the 2005 APEC confer- enceand the level of technology in Korea was publicized to all nations. E. HUBO FX-1 (2006) HUBO FX-1 is a practical biped robot that can carry a person. It has a height of 139 [cm] (199 [cm] including the cockpit), a weight of 120 [kg] (150 [kg] including its body covers and a cockpit), and 12 dof. This giant Fig. 3 KHR-2 and specifications [4] robot was designed to sufficient payload while it walks dynamically carrying one passenger. As its payload ca- The DOF was increased to 41 because of the hands and pacity is 100 [kg], an average person is able to ride on head. The operating system was changed to Windows HUBO FX-1 easily. As joint actuators, AC servo mo- XP with RTX for real time control. Fig. 1 shows the tors and harmonic reduction gears were used to generate overall system configuration of KHR-2. The subsequent sufficient torque and power as well as to minimize the versions of humanoid robots in the HUBO Laboratory backlash. continued to use this same configuration. F. HUBO 2 (2009) C. HUBO (KHR-3) (2005) HUBO2 is the latest version of the HUBO series. This KHR-3 was the third version of the KHR series and robot was developed in 2009. Based on our experi- we named this version HUBO. This robot was developed ence over 10 years, we accomplished an improved robot in 2004. Most of the specifications of KHR-3 were sim- system and performance ability in such tasks as mo- ilar to KHR-2. However, mechanical stiffness of links tion, walking and even running. The main goal of the and reduction gear capacity of the joints were modified HUBO2 design was to achieve the lightest human-size and improved. HUBO can dance with various motions humanoid robot in the world. The design for the light and can do sign language, moving five fingers indepen- arms of HUBO 2 changed to 7-DOF and became more

JRSJ Vol. 30 No. 4 —34— May, 2012 Development of Humanoid Robots in HUBO Laboratory, KAIST 369

Fig. 4 HUBO (KHR-3) and speciﬁcations [3]

Fig. 7 HUBO 2

Table 1 Control schemse and controller of HUBO series

Fig. 5 Albert HUBO and speciﬁcations [8]

ing pattern generator with a real-time feedback controller, bipedal walking of our humanoid robots is accomplished. A. Walking Patterns In the early stages of developing humanoid robots in Fig. 6 HUBO FX-1 and specifications [13] our laboratory, we only had technologies for motor control, not the humanoid robot, so a gait trajectory was compact so that its motion could be swift. Through designed offline very heuristically at the beginning of its light weight, HUBO 2 can run at a maximum speed the KHR series [2] [3]. However, as the robot was up- of 3.6 [km/h] [18] [19]. Moreover, new the walking al- graded, the walking pattern was modified with respect gorithm permitted stretched leg walking [20] [21], which to a stability index like ZMP. We performed experi- was different from previous robots. ments with those patterns and verified that our walking pattern was very simple and also well designed for 3. Walking and Control walking [5] [6]. Moreover, to change the robot walking Walking is the essential requirement of a humanoid pattern in real time, we made an online walking pattern robot. We have been exploring how the robot can walk generator using simple functions and walking parame- stably and how we can control the robot. The walk- ters [9] [10]. That is, the gait trajectory functions gen- ing algorithm of the HUBO series developed until now erated the relative position trajectories of the two feet consists of two categories: the walking pattern and the with respect to the pelvis center using a few parameters real-time feedback control. By using a simple walk- in real time. To find good paramerters of walking, we

日本ロボット学会誌 30 巻 4 号 —35— 2012 年 5 月 370 Jung-Woo Heo In-Ho Lee Jun-Ho Oh also tried to use reinforcement learning [23]. applied to the other mobility skills of the robot, for ex- B. Real-time Feedback Control ample, climbing stairs, running, etc. From the viewpoint of stability, even a well-designed 4. Conclusion walking pattern cannot prevent the robot from falling down. We didn’t use a stability index to make the walk- We have performed pioneer humanoid robot research. ing pattern, so our walking pattern did not guarantee We have made mechanical progress in creating hu- stability. Moreover, as a result of unexpected exter- manoid robots that have light weight, high mobility and nal disturbance or environmental changes, for example, a strong mechanical design. We have achieved success large upper body motions, vibrations of the body parts, in such areas of dynamic control as walking algorithms, uneven ﬂoors, etc., the robot walking pattern could walking controllers and running patterns in for bipedal be exceeded over stability region. Therefore, real-time robots. feedback control is essential to maintaining dynamic Nowaday, other research institutes are also achieving balance in real time considering unexpected factors. successes, so what is next for the humanoid robot? The real-time feedback control can be categorized in We expect to see more advanced humanoid robots, three groups of the scheme depending on the control ob- which will come to us in the near future. Until that day jective: balancing control, posture control and uneven comes, we will keep struggling to develop the ultimate terrain control. This order is the same as the time se- humanoid robot that people dream of. quence in which we developed controllers as necessary. References First of all, we used balancing control [2] [3] [6] that involves self balancing during walking. This controller [ 1 ] J.-H. Kim, S.-W. Park, I.-W. Park and J.-H. Oh: “Development of a Humanoid Biped Walking Robot Plaform KHR-1—Initial compensates for the un-modeled dynamics of the walk- Design and Its Performance Evaluation,” Proceedings of The ing pattern by using a damping controller and a ZMP Third IARP International Workshop on Humanoid and Human Friendly Robotics, pp.14–21, 2002. compensator. These controllers play the most impor- [ 2 ] J.-H. Kim and J.-H. Oh: “Torque feedback control of the hu- tant role for not only stable walking but also for bal- manoid platform KHR-1,” IEEE-RAS Int. Conf. on Humanoid anced standing itself. Robots, 2003. [ 3 ] J.-H. Kim and J.-H. Oh: “Realization of dynamic walking for Secondly, posture control [22] is involved in keeping the humanoid robot platform KHR-1,” Journal of Advanced the robot’s required posture. For example, the vibra- Robotics, vol.18, no.7, pp.749–768, 2004. [ 4 ] I.-W. Park, J.-Y. Kim, S.-W. Park and J.-H. Oh: “Develop- tion reduction controller maintains the swing leg posi- ment of humanoid Robot platform KHR-2(KAIST Humanoid tion by reducing the vibration of the swing leg while Robot-2),” International Journal of Humanoid Robotics, vol.2, no.4, pp.519–536, 2005. the robot is walking. Vibration is a problem that easily [ 5 ] J.-Y. Kim, I.-W. Park, J.-H. Lee, M.-S. Kim, B.-K. Cho and occurs because the robot is not a rigid body. Posture J.-H. Oh: “System Design and Dynamic Walking of Humanoid control is helpful in reducing unexpected factors and Robot KHR-2,” ICRA 2005, pp.1443–1448, 2005. [ 6 ] J.Y. Kim, I.W. Park and J.H. Oh: “Experimental realization maintaining an accurate position. of dynamic walking of the biped humanoid robot KHR-2 using Finally, robot is controlled with concernig about un- zero moment point feedback and inertial measurement,” Adv. Robot. vol.20, no.6, pp.707–736, 2006. even terrain walking [16]. Real terrains that robots need [ 7 ] J.-Y. Kim, I.-Woo, P. Jung, H. Lee and J.-H. Oh: “Experi- to walk are rarely perfectly even. When a robot unex- ments of Vision Guided Walking of Humanoid Robot, KHR-2,” Humanoid 2005, pp.135–140, 2005. pectedly steps on uneven terrain, the robot tilts to the [ 8 ] I.-W. Park, J.-Y. Kim, J. Lee and J.-H. Oh: “Mechanical outside or inside. This means that the robot cannot Design of the Humanoid Robot Platform, HUBO,” Advanced step on the position and at the time that we designed Robotics, vol.21, no.11, pp.1305–1322, 2007. [ 9 ] I.-W. Park, J.-Y. Kim, J. Lee and J.-H. Oh: “Online Free Walk- it to do. Therefore we control the landing moment that ing Trajectory Generation for Biped Humanoid Robot KHR-3 steps on locally uneven terrain. We also make the robot (HUBO),” IEEE Int. Conf. on Robotics & Automation, Serial.1, pp.1231–1236, 2006. maintain an upright pose when the robot steps on a [10] I.-W. Park, J.-Y. Kim and J.H. Oh: “Online Walking Pattern globally uneven terrain. Generation and Its Application to a Biped Humanoid Robot— KHR-3(HUBO),” Advanced Robotics, vol.22, no.2–3, pp.159– Using these kinds of controllers, recently, the HUBO 190, 2008. series robots can walk stably at a speed of 1.5 [km/h], [11] J.-H. Oh, D. Hanson, W.-S. Kim, I.-Y. Han, J.-Y. Kim and and even walking on uneven terrain they are success- I.-W. Park: “Design of android type humanoid robot albert hubo,” Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst., ful. Moreover, these real-time feedback controllers are pp.1428–1433, 2006.

JRSJ Vol. 30 No. 4 —36— May, 2012 Development of Humanoid Robots in HUBO Laboratory, KAIST 371

[12] I.-W. Park, J.-Y. Kim, B.-K. Cho and J.-H. Oh: “Control hard- manoid Robots, Serial.1, 2007. ware integration of a biped humanoid robot with an android [18] B.-K. Cho and J.-H. Oh: “Running Pattern Generation with head,” Robotics and Autonomous Systems, vol.56, no.1, pp.95– a Fixed Point in a 2D Planar Biped,” International Journal of 103, 2008. Humanoid Robotics, vol.6, Issue 2, pp.241–264, 2009. [13] J. Lee, J.Y. Kim, I.W. Park, B.K. Cho, M.S. Kim, M.I. Kim and [19] B.-K. Cho, S.-S. Park and J.-H. Oh: “Controllers for Run- J.H. Oh: “Development of a humanoid robot platform HUBO ning in the Humanoid Robot, HUBO,” 2009 IEEE-RAS Inter- FX-1,” SICE-ICASE Int. Joint Conf ’06, pp.1190–1194, 2006. national Conference on Humanoid Robots, 2009. [14] J.-Y. Kim, J. Lee and J.-H. Oh: “Experimental Realization of [20] M.-S. Kim, I.-W. Park, J.-Y. Kim and J.-H. Oh: “Stretch- Dynamic Walking for a Human-Riding Biped Robot, HUBO Legged Walking in Sagittal Plane,” IEEE-RAS International FX-1,” Advanced Robotics, vol.21, no.3–4, pp.461–484, 2007. Conference of Humanoid Robots, Serial.1, 2007. [15]J.-Y.Kim,J.Lee,I.-W.ParkandJ.-H.Oh:“VibrationReduc- [21] M.-S. Kim, I.-H. Kim, S.-S. Park and J.-H. Oh: “Realization of tion Control For Human-Riding Biped Robot HUBO FX-1,” Stretch-legged Walking of the Humanoid Robot,” 2008 IEEE- 4th IFAC Symposium on Mechatronic Systems, Mechatronics RAS International Conference on Humanoid Robots, 2008. 2006, Serial.1, pp.638–643, 2006. [22] M.-S. Kim and J.-H. Oh: “Posture Control of a Humanoid [16] J.-Y. Kim, I.-W. Park, J.-H. Oh: “Walking Control Algorithm Robot with a Compliant Ankle Joint,” International Journal of Biped Humanoid Robot on Uneven and Inclined Floor,” of Humanoid Robotics, vol.7, no.1, 2010. Journal of Intelligent Robot System, vol.48, pp.457–484, 2007. [23] J.-H. Lee and J.H. Oh: “Biped Walking Pattern Generation [17] B.-K. Cho and J.-H. Oh: “Controller Design and Experimen- Using Reinforcement Learning,” International Journal of Hu- tal Approach on the Dynamic Walking on the Spot in Pla- manoid Robotics, vol.6, no.1, 2009. nar Biped Robot,” IEEE-RAS International Conference of Hu-

Jung-Woo Heo In-Ho Lee received his BS degree in Mechanical En- received his BS and MS degree in Me- gineering from Hanyang University, Seoul, chanical Engineering from Korea Ad- Korea, in 2007, MS degree in Mechanical vanced Institute of Science and Technology Engineering from Korea Advanced Insti- (KAIST), Dae-jeon, Korea, in 2009, 2011. tute of Science and Technology (KAIST), Since 2011, he has been a Ph.D candidate Dae-jeon, Korea, in 2009. Since 2009, he at the KAIST and worked on the project has been a Ph.D candidate at the KAIST and worked on of HUBO. His research interests include design and control the project of HUBO. His research interests include motion of biped humanoid robot and motor control. control and bipedal walking control of humanoid robot.

Jun-Ho Oh received his BS and MS degrees in Mechan- ical Engineering from Yonsei University, Seoul, South Korea, and has PhD degree in Mechanical Engineering from University of California, Berkeley, in 1977, 1979, and 1985 respectively. He was a Researcher with the Korea Atomic Energy Research Institute, from 1979 to 1981. Since 1985, he has been with the Department of Mechanical Engineering, KAIST, where he is currently a professor and a director of Humanoid Robot Research Cen- ter. He was a Visiting Research Scientist in the University of Texas Austin, from 1996 to 1997. His research interests include humanoid robots, adaptive control, intelligent control, nonlinear control, biomechanics, sensors, actuators, and Posture Control of a Humanoid Robot with a Compli- ant Ankle Joint 29 applications of micro processor. Dr. Oh is a member of the IEEE, KSME, KSPE and ICASE.

日本ロボット学会誌 30 巻 4 号 —37— 2012 年 5 月